gist cost estimator failing

This might be gist-related or not. The cost estimator
is wrongly estimating an higher cost on Index Scan then
it estimates on Seq Scan (while reality shows the difference).

I've found out that the gistcostestimate function is just
a stub calling the genericcostestimate, can this be the problem ?

Can you point me to where to inspect further ?

Seq Scan on plmshp02_1
(cost=0.00..19.14 rows=26 width=81)
(actual time=0.54..61.68 rows=75 loops=1)
Total runtime: 61.83 msec

Index Scan using plmshp02_1_gist on plmshp02_1
(cost=0.00..81.53 rows=26 width=81)
(actual time=0.39..19.22 rows=75 loop=1)
Total runtime: 19.37 msecs

--strk;